elbe key features: LBM and GPUs
The elbe code uses a variety of one-, two- and three-dimensional Lattice-Boltzmann models on the basis of D1Q3, D2Q9, D3Q19 and D3Q27 stencils. Available models are LBM singlephase (2D, 3D), multiphase (2D, 3D), free surface (2D, 3D) and shallow water (1D, 2D). On top of the conventional, well-known SRT and MRT collision operators, advanced Cascaded and Cumulant LB collision operators are available.
Turbulence is modeled with a Smagorinsky subgrid model, in which the effects of the non-resolved eddies is represented by an additional space-varying turbulent viscosity. The model was validated with state-of-the-art turbulence benchmarks. Moreover, dynamic Smagorinsky models and implicit LBM-LES models are available.
Surface mounted cube test case.
The elbe code is accelerated with the help of graphics processing units (GPUs) and the NVIDIA CUDA framework. Apart from several GPU-equipped workstations that are used for code development, small-scale parameter studies and real-time simulations with online visualization, the elbe team operates a small multi-GPU system with C2075 (hardware donation from NVIDIA) and K40 boards, mainly used for medium-sized parameter studies and Multi-GPU runs. On top, the team has access to professional compute servers in the TUHH compute center (12x K80 GPUs) and at the University of Hamburg (54x K80 GPUs) for large-scale studies with high spatial and temporal resolution.For live demos and simulations at exhibitions and fairs, elbe2go, a portable Mini-ITX machine with a Titan X GPU that was tailor-made for local simulations on the desktop, is used.
The team is supported by NVIDIA, as part of the NVIDIA's Academic Partnership Programm (APP), for the research on Fast Efficient Numerical Simulations of Violent Free-Surface Flows using CUDA. Moreover, TUHH and the elbe team was awarded with the NVIDIA CUDA Research Center Status (CRC, October 2014).
The efficient LBM bulk scheme is combined with accurate free-surface capturing techniques, a fluid-structure interaction interface and tailor-made algorithms for GPU-accelerated grid generation and online visualization.
- Scrutinizing Lattice Boltzmann Methods for Direct Numerical Simulations of Turbulent Channel Flows.
M. Gehrke, C.F. Janßen and T. Rung, Computers & Fluids 156:247-263, October 2017